| Since the last century,there have been some scholars devoted to the rapid diagnosis of crop diseases.Scholars hope to identify and even control crop diseases by using electronic devices instead of manual observation.In recent years,with the development of deep learning,the application of deep learning for crop disease identification and detection has become a more feasible and low-cost method.Image processing technology and machine learning methods require manual extraction of features.To solve this problem,this paper proposes a deep learning-based navel orange leaf disease recognition method.In the previous methods of using deep learning to identify crop diseases,the data set was taken under a single background in the laboratory,and this article established a complex background navel orange leaf image data set containing six common navel orange diseases.Using deep learning networks to train and learn these images can effectively identify and classify crop diseases.This article takes the economic crop navel orange disease in southern Jiangxi as the research object and launches the classification algorithm research.We use the Mobile Net V2 model as the main network and compare it with other network models in terms of recognition speed,model scale,and accuracy.The results show that the use of the method based on the Mobile Net V2 model reduces the consumption of prediction time and the size of the model,while maintaining good classification accuracy.Finally,we designed to use Mobile Net V2 to develop mobile terminals to help identify agricultural diseases.The main contents of this article are:(1)In the field of navel orange diseases,there is currently no publicly available high-quality image data set.For this reason,this research has constructed a picture data set of common diseases of navel oranges.This data set is used in the research of this article and can be further published to promote the development of the field of navel orange disease identification.In the past,most of the research on plant disease identification was obtained by collecting plant leaves and shooting in a single background in the laboratory.The single background of diseased leaves ignores the complex background in real applications.This time the picture data is taken on the spot in the orchard.As a result,the data set is more authentic.(2)In the process of studying the classification algorithm,the Inception Net V3 network model of the classic Goog Le Net series is used for migration learning experiments,and the trained model has a high accuracy in recognition and detection training.In order to achieve deployment on mobile devices,this paper also carried out a research on the classification of navel orange diseases based on the Mobile Net V2 network model.The comparison of the two models shows that the accuracy of the Inception Net v3 model is larger,but the model memory is larger,and when the accuracy of Mobile Net v2 is similar,the model memory is smaller.(3)Finally,this paper designs a navel orange leaf disease detection system based on the Mobile Net V2 model,and develops a mobile phone application APP that can quickly identify the disease. |
